Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Generalised Exponentiated Gradient Approach to Enhance Fairness in Binary and Multi-class Classification Tasks

Maryam Boubekraoui, Giordano d'Aloisio, Antinisca Di Marco

Abstract

The widespread use of AI and ML models in sensitive areas raises significant concerns about fairness. While the research community has introduced various methods for bias mitigation in binary classification tasks, the issue remains under-explored in multi-class classification settings. To address this limitation, in this paper, we first formulate the problem of fair learning in multi-class classification as a multi-objective problem between effectiveness (i.e., prediction correctness) and multiple linear fairness constraints. Next, we propose a Generalised Exponentiated Gradient (GEG) algorithm to solve this task. GEG is an in-processing algorithm that enhances fairness in binary and multi-class classification settings under multiple fairness definitions. We conduct an extensive empirical evaluation of GEG against six baselines across seven multi-class and three binary datasets, using four widely adopted effectiveness metrics and three fairness definitions. GEG overcomes existing baselines, with fairness improvements up to 92% and a decrease in accuracy up to 14%.

A Generalised Exponentiated Gradient Approach to Enhance Fairness in Binary and Multi-class Classification Tasks

Abstract

The widespread use of AI and ML models in sensitive areas raises significant concerns about fairness. While the research community has introduced various methods for bias mitigation in binary classification tasks, the issue remains under-explored in multi-class classification settings. To address this limitation, in this paper, we first formulate the problem of fair learning in multi-class classification as a multi-objective problem between effectiveness (i.e., prediction correctness) and multiple linear fairness constraints. Next, we propose a Generalised Exponentiated Gradient (GEG) algorithm to solve this task. GEG is an in-processing algorithm that enhances fairness in binary and multi-class classification settings under multiple fairness definitions. We conduct an extensive empirical evaluation of GEG against six baselines across seven multi-class and three binary datasets, using four widely adopted effectiveness metrics and three fairness definitions. GEG overcomes existing baselines, with fairness improvements up to 92% and a decrease in accuracy up to 14%.
Paper Structure (39 sections, 1 theorem, 68 equations, 6 figures, 7 tables, 1 algorithm)

This paper contains 39 sections, 1 theorem, 68 equations, 6 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Research Objectives
  3. Main Contributions
  4. Roadmap
  5. Background and Related Work
  6. Background Knowledge on Bias Assessment and Mitigation
  7. Related Work
  8. Exponentiated Gradient Approach
  9. Methodology
  10. Fairness Constraints
  11. Fairness Constraints as Linear Moment Conditions
  12. Generalized Exponentiated Gradient (GEG)
  13. Convergence Analysis
  14. Implementation Details
  15. Evaluation
  16. ...and 24 more sections

Key Result

Theorem 1

Let $\eta = \Theta(1/\sqrt{T})$ be the learning rate used in the Exponentiated Gradient updates. Under the boundedness assumptions and a $\tau$-approximate cost-sensitive oracle, the iterate $(Q_T,\boldsymbol{\lambda}^{(T)})$ produced by GEG satisfies In particular, $Q_T$ is an $O(1/\sqrt{T})$-approximate solution to eq15, up to $\tau$.

Figures (6)

  • Figure 1: Experimental process
  • Figure 2: RQ$_1$: Pareto optimality of baseline and GEG variations considering each pair of effectiveness and fairness metrics.
  • Figure 3: RQ$_2$: Pareto optimality of base LR, base EG approaches, and GEG-CP considering each pair of effectiveness and fairness metrics for binary classification.
  • Figure 4: RQ$_3$: Pareto optimality of DEMV, Blackbox and GEG variations considering each pair of effectiveness and fairness metrics for multi-class classification.
  • Figure 5: RQ$_4$: Pareto optimality of RF and GEG variations considering each pair of effectiveness and fairness metrics.
  • ...and 1 more figures

Theorems & Definitions (6)

  • Definition 1: Multi-class Demographic Parity (DP)
  • Definition 2: Multi-class Equalized Odds (EO)
  • Definition 3: Positive Label Demographic Parity
  • Definition 4: Positive Label Equalized Odds
  • Theorem 1: Convergence of GEG
  • proof